Brown, A.J., Thomas, M.
, "Reengineering the Naval Ship Concept Design Process", From Research to Reality in Ship Systems Engineering Sy m-
posium, ASNE, September, 1998.
Reengineering the Naval Ship Concept Design Process
Dr. Alan Brown, Virginia Tech and LCDR Mark Thomas, USN
Abstract
Naval ship concept design is very much an ad
hoc process. Selection of design concepts for as-
sessment is guided primarily by experie nce, design
lanes, rules-of-thumb, preference and imagination.
Objective attributes are not adequately synthesized
or presented to support efficient and effective deci-
sions. Attributes are often qualitative, inconsistent,
and not provided to design engineers in a format they
can use. The design space is very large, non-linear,
discontinuous, and bounded by a variety of con-
straints and thresholds. These problems make a
structured search of design space difficult. Without
a structured search, there is no rational way to
measure the optimality of selected concepts relative
to the millions of other concepts that have not been
considered or assessed. Responsible decisions can-
not be made without this information and perspec-
tive.
This paper addresses these problems in the con-
text of a systems approach to naval ship concept de-
sign. Multiattribute value theory (MAVT) and the
Analytical Hierarchy Process (AHP) are used to
synthesize an effectiveness function. A Pareto Ge-
netic Algorithm (PGA) searches design parameter
space and identifies non-dominated design concepts
in terms of cost and effectiveness. Design concepts
are presented graphically as points on a non-
dominated cost-effectiveness frontier for considera-
tion by decision-makers. A simplified surface-
combatant design demonstrates the process.
Introduction
Despite steady improvement in design tools, and
excellent progress in concurrent and systems engi-
neering (Kramer, 1996, and Tibbitts, 1995), naval
ship concept design is still very much an ad hoc
process. Elements missing from this process are:
A quantitative methodology for synthesizing
a manageable set of critical, but dissimilar
objective attributes
An efficient method to search design space
for non-dominated concepts based on these
attributes
An effective format to present these non-
dominated concepts for rational selection
Critical naval ship objective attributes are mission
effectiveness, cost and risk. Each of these overall
attributes includes a number of specific attributes or
measures such as mission-specific Measures of Ef-
fectiveness (MOEs) whose cumulative value must
be synthesized in the overall measure.
Effectiveness, cost and risk are dissimilar attrib-
utes, and require different units of measure. They
cannot rationally be combined into a single objective
attribute. They must be presented individually, but
simultaneously in a manageable format for tradeoff
and decision-making. Manageable implies that only a
limited number of attributes can be considered simul-
taneously. This requires either looking at one piece
of the problem at a time, or combining similar obje c-
tive attributes into an overall measure or index.
Effectiveness and risk are relatively abstract ob-
jectives that are sometimes difficult to measure
quantitatively. The effectiveness of a few concepts
can be analyzed using war gaming and other com-
plex models, but this approach is not practical when
evaluating many concepts in a structured search of
design space. This paper presents a methodology for
calculating an Overall Measure of Effectiveness
(OMOE) index using expert opinion to synthesize di-
verse inputs such as defense guidance, mission re-
quirements, threat, war game results and experience.
Risk requires a similar treatment, and will be ad-
dressed in subsequent work.
MOEs describe mission effectiveness in specific
scenarios. Examples of MOEs are conflict duration,
territory lost or gained, casualties, and targets de-
stroyed. Measures of Performance (MOPs) define
the performance of the ship system independent of
mission scenarios. Examples of MOPs are sustained
speed, endurance and signatures. Design parame-
ters (DPs) provide the physical description of the
ship system. DPs determine MOPs, and MOPs Figure 1 illustrates this concept for a simple two-
determine MOEs. DPs also determine cost and risk. objective (cost-effectiveness) problem. The heavy
Ultimately, a ship design is defined by specifying curve represents non-dominated solutions or the Pa-
millions of DPs, in thousands of drawings, and with reto-optimal frontier. The preferred design should
libraries full of technical specifications and informa- always be one of these non-dominated solutions. Its
tion. Even in its simplest concept form, the definition selection depends on the decision-makers prefer-
of a balanced ship design requires many DPs. The ence for cost and effectiveness. This preference
functional relationship of these DPs cannot be de- may be affected by the shape of the frontier and
scribed in a closed-form set of equations. cannot be rationally determined a priori.
A total-system approach to ship design makes an
already complex problem more complex. The goal Effectiveness
of a total system approach is to optimize the life cy-
cle cost-risk-effectiveness of the total ship system. Non-dominated
This system includes the ship and everything outside Solutions
the ship that either affects it or is affected by it. It
usually requires an iterative and interactive process
that depends on an effective concurrent engineering Feasible
organization to produce a true total-system result. Region
The hierarchy of systems and subsystems ni -
cluded in a total-ship-system is rightly called a "su-
persystem" (Hockberger, 1996). At the bottom of
this hierarchy are the detailed components and char-
Cost
acteristics that define the ship. Many lower-level Figure 1. Two-Objective Attribute Space
system decisions can be made at their own level or
one higher. Others must be determined at the total
ship level. Some compromise between global and lo- OMOE
cal optimization is essential to keep the problem 1.0
0.9
manageable. The number of DPs at any level must 0.8
be kept to the minimum necessary to capture impor- 0.7
tant interdependence. The highest level of optimiza- 0.6
0.5
tion should consider only those variables that have a 0.4
major impact on ship balance. Frequently combat 0.3
high
systems, HM&E systems and ship characteristics 0.2
0.1
can be grouped into synergistic packages or suites. 0.0 med
This reduces the number of variables that must be 1.0
1.5 Risk Index
2.0
managed early in the design process. 2.5
3.0
low
LCC ($B) 3.5
The primary objective of the concept design
process (as defined here) is to identify non-
Figure 2. Three-Objective Attribute Space
dominated and feasible concepts for selection by de-
cision-makers based on the objective attributes of When considering three attributes, the non-
cost, effectiveness and risk. Ideally, there should be dominated frontier is a surface, as illustrated in Fig-
no bias or preference for particular DPs or MOPs. ure 2. Points on this surface represent feasible ships,
They are only fundamental and intermediate parame- and can be mapped to specific design parameters.
ters. Cost, effectiveness and risk are the relevant With such a surface, the full range of cost-risk-
objective attributes. effectiveness possibilities can be presented to deci-
A non-dominated solution, for a given problem sion-makers, knees in the curve can be seen
and constraints, is a feasible solution for which no graphically, trade-off decisions can be made, and
other feasible solution exists which is better in one specific design concepts can be chosen for further
objective attribute and at least as good in all others. analysis.

2
 Expert Opinion
Mission M&S
Mission Need Engineering Process /
Statement (MNS) Analysis LCC Model Schedule
Determine M&S Model
Mission Need Concept
Studies Zigzag down hierarchies
Approval

Design Synthesis and Analysis

Exploratory Define and Relate to Cost-Effectiveness-Risk Surface Build
Mission calculate OMOE Functional OMOE = f($, risk) Strategy
Analysis OMOE = g(MOP's) Requirements Product Model PV's
CR's FR's DP's

EXPLORATION MISSION DOMAIN FUNCTIONAL PHYSICAL PROCESS

DOMAIN DOMAIN DOMAIN

Constraints
R&D and
Exploratory
Design Define "possibilities" as function
of technology risk.

Figure 3. Notional Concept Design Process

Proposed Design Process Framework deficiencies, and estimates the force level, mix and
mission performance required of future ships.
It is helpful to think of the design process as a
Ship
sequential mapping between four domains as shown
in Figure 3: 1) the mission or customer domain; 2) the Mission /
Customer
Functional Physical Process/
Build Strategy

functional domain; 3) the physical domain; and 4) the Warfare Enclose and protect Structure Max Lift Weight
Mobility Max Dimensions
process domain (Suh, 1990). Decisions made in Sustainability
Vulnerability
Keep out sea
Control flooding
Hullform
Shell
Plate curvature
Catalog of Shapes

each domain are mapped into the subsequent domain, Susceptability

Reliability
Partition volume
Provide deck shelter
Watertight Tbhd
Decks
Block size
Nuclear/non-nuclear
Flexibility Protect Tanks
moving from what to how in each mapping, and Provide mobility Deckhouse
Propulsion

then zigzagging down hierarchies in each domain as Provide stable platform

Produce F&A Thrust Main engines
Sustain Reduction gear
the design is defined in increasing detail. Customer Replenish
Maneuver
Shafting
Propeller

requirements (CRs) are the what for the functional Provide electric power Control system
Steering
Generate Electric Power
domain, functional requirements (FRs) are the Transmit
Convert Generators
how. FRs are the what for the physical domain, Support Cables
Power conversion
Auxiliary Systems
design parameters (DPs) are the how. DPs are Ventilate
Cool
Provide habitability Ventilation system
the what for the process domain, process variables Provide safety/damage control
Fight/Support
Chilled water system
Crew support systems
DC systems
(PVs) are the how. Results are fed back to the AAW Payload
NSFS
customer, CRs are refined, and the mapping contin- ASUW
ASW
Sensors
Weapons

ues. A notional top-level design hierarchy consistent C4I

SEW
Communications
ECM
FSO
with this scheme is shown in Figure 4. STW

Figure 4. Notional Top Level Design Hierarchy

Exploration. The naval ship design process is
initiated with the definition of a mission need and di-
Important products of exploratory mission analy-
rection to begin concept studies, but signific ant effort
sis include the Mission Need Statement (MNS),
precedes this direction.
threat, required operational capabilities (ROCs), pro-
Exploratory mission analysis provides a descrip-
jected operational environment (POE), mission sce-
tion of projected world political/military environments,
narios and MOEs. Also important are war-fighting
and maintains a time-phased assessment of the
model validation and sensitivity analysis including the
threat and future mission scenarios consistent with
identification of critical ship MOPs.
defense planning guidance. It determines joint force
Research, development, technology demonstra-
and naval force structure, identifies future mission
tion, and exploratory design identify, develop and

3
evaluate new systems, technology, and ship con-
cepts. Together with proven systems and concepts
these pieces must be assembled into balanced, feasi-
ble, and cost-effective ship designs that must satisfy
the evolving mission need in an evolving environ-
ment. Products of R&D and Exploratory Design in-
clude new technology and systems, ship concepts,
and a preliminary definition of feasible ship and sys-
tem performance. Expected performance is de-
scribed using upper and lower bounds on MOPs.
These bounds are a function of technology risk that
may be quantified and included as a system attribute.
Mission Domain. Customer needs and re-
quirements are determined, and effectiveness is as-
sessed in the Mission Domain. Initial inputs to this
domain are the MNS, POE, threat definition and mis-
sion scenarios. Customer requirements (CRs) are
the output. In this research, CRs are specified in
three ways: 1) required operational capabilities
(ROCs); 2) performance constraints, goals and
thresholds; and 3) an overall mission measure of ef-
fectiveness (OMOE) index which defines mission ef-
fectiveness as a function of ship MOPs. The defini-
tion of a quantitative relationship between mission ef-
fectiveness and ship performance in the mission do-
main is an essential prerequisite to a disciplined
search of design parameters in the physical domain.
Modeling, simulation and expert opinion are used to
define the OMOE as a function of MOPs. Some
MOPs are binary, the ship has a capability or it
doesn't. Others are continuous, such as sustained
speed or endurance.
Functional Domain. In the Functional Domain,
top level functional requirements might include en-
close and protect, provide mobility, support and fight.
Consideration of functional requirements is most im-
portant when new design solutions or technologies
are sought and when paradigms are reevaluated.
Physical Domain. Design parameters define
the ship in the Physical Domain. The selection, syn-
thesis and balance of DPs determine ship MOPs,
and ultimately determine mission effectiveness. Cost
and risk are determined as a function of DPs and
process variables (PVs).
In this research, a simple ship-synthesis model is
used to synthesize and balance designs in the physi-
cal domain, and to calculate the first level of ship
MOPs. Balance requires that physical and func-
tional constraints are satisfied. A genetic algorithm
4
(GA) is used to search DP space in the physical do-
main, and to generate and identify concepts on the
non-dominated objective attribute frontier. More so-
phisticated tools, models and simulations can be used
later in the design process on selected concepts to
refine the designs, demonstrate feasibility, and m i -
prove MOP calculations. Analysis results are added
to a design knowledge base and applied to update
model parametric equations, MOP, cost and risk cal-
culations. This provides a dynamic landscape or en-
vironment for the genetic algorithm over the course
of the design process. The updated non-dominated
frontier is used to reevaluate and adjust earlier top
level DP decisions during the design process until
further design changes are no longer cost-effective.
Process Domain. Critical design parameters
must be related to process variables (PVs). These
process variables may be synthesized in a build strat-
egy. The build process must be refined at each level
of the design hierarchy to insure feasibility and
maximize producibility. PVs effect cost and risk.
Building the OMOE Function
Early in the design process, designers and engi-
neers require a working model to quantify operators
and policy-makers definition of mission effective-
ness, and define its functional relationship to ship
MOPs. This quantitative assessment of effective-
ness is fundamental to a structured optimization
process.
There are a number of inputs which must be in-
tegrated when determining overall mission effective-
ness: 1) defense policy and goals; 2) threat; 3) exist-
ing force structure; 4) mission need; 5) mission sce-
narios; 6) modeling and simulation or war gaming re-
sults; and 7) expert opinion. Ideally, all knowledge
about the problem could be included in a master war-
gaming model to predict resulting measures of effec-
tiveness for a matrix of MOP inputs in a series of
probabilistic scenarios. Regression analysis could be
applied to the results to define a mathematical rela-
tionship between input ship MOPs and output MOEs.
The accuracy of such a simulation depends on mod-
eling the detailed interactions of a complex human
and physical system and its response to a broad
range of quantitative and qualitative variables and
conditions including ship MOPs. Many of the inputs
and responses are probabilistic so a statistically sig-
nificant number of full simulations must be made for
each set of discrete input variables. This extensive
modeling capability does not yet exist for practical
applications.
An alternative to modeling and simulation is to
use expert opinion directly to integrate these diverse
inputs, and assess the value or utility of ship MOPs in
an OMOE function. This can be structured as a
multi-attribute decision problem. Two methods for
structuring these problems dominate the literature:
Multi-Attribute Utility Theory (MAUT, Keeney and
Raiffa, 1976) and the Analytical Hierarchy Process
(AHP, Saaty, 1996). In the past, supporters of these
theories have been critical of each other, but recently
there have been efforts to identify similarities and
blend the best of both for application in Multi-
Attribute Value (MAV) functions (Belton, 1986).
This approach is adapted here for deriving an
OMOE.
The analytical hierarchy process (AHP) is a tool
developed by Saaty (1996) for solving multi-attribute
decision problems. It uses a hierarchical structure to
abstract, decompose, organize and control the com-
plexity of decisions involving many attributes, and it
uses informed judgment or expert opinion to measure
the relative value or contribution of these attributes
and synthesize a solution. Pair-wise comparison and
a maximum eigenvalue approach extract and quan-
tify this relative value. The method allows and
measures inconsistency in value measurement, and is
able to consider quantitative and qualitative attrib-
utes.
A hierarchy is a simplified abstraction of the
structure of a system used to study and capture the
functional interactions of its attributes, and their im-
pact on total system behavior or performance. It is
based on the assumption that important system enti-
ties or attributes, which must first be identified, can
be grouped into sets, with the entities of one group or
level influencing the entities of the neighboring group
or level. One can conceptualize a hierarchy as a
bottoms-up synthesis of influence on the top level
behavior of a system, or as the top down distribution
of influence of top level behavior to low level attrib-
utes. Alternatives are compared in terms of the
lowest level attributes and this comparison is rolled
up through hierarchy levels to an assessment of rela-
tive overall system behavior or performance.
5
The first step in building an AHP hierarchy is to
identify critical attributes affecting the decision or
system behavior. The level of detail of these attrib-
utes depends on the decision being made. These at-
tributes are then organized into a hierarchy structure
that follows a logical breakdown or categorization as
shown in Figure 5. System options or alternatives
comprise the bottom hierarchy level.
Next, the relative influence of each attribute on
system performance and attribute values for each al-
ternative must be estimated. Saaty recommends a
nine level dominance scale for the pair-wise com-
parison of attribute influence on higher level attrib-
utes. This results in a ratio scale comparison of at-
tributes. Pair-wise comparison or cardinal values
may be used to assign attribute values for each alter-
native. Pair-wise comparison generates more in-
formation than is necessary with individual absolute
measurements or estimates. The AHP synthesizes
and evaluates the consistency of this redundant in-
formation and calc ulates best-fit relative values.
Although the AHP was developed primarily for
comparison of management alternatives, it has also
proven to be a robust method for application in
MAVT (Belton, 1986). The AHP provides a struc-
tured method for deriving an additive weighted value
function, and by careful application can also be used
to derive non-linear attribute value or utility without
the more cumbersome lottery comparison approach.
The OMOE function must include all important
effectiveness/performance attributes, both discrete
and continuous, and ultimately be used to assess an
unlimited number of ship alternatives. Successful
application AHP/MAVT to this problem requires a
very structured and disciplined process as follows:
1. Identify, define and bound decision at-
tributes. Identify critical mission scenarios. Iden-
tify MOE(s) for each mission scenario. Establish
goals and thresholds for all MOEs. Identify ship
MOPs critical to mission scenario MOE assessment
and consistent with the current design hierarchy
level. Set goals and thresholds for these MOPs.
2. Build OMOE/MOP hierarchy. Organize
MOEs and MOPs into a hierarchy as shown in Fig-
ure 5, with specific ship MOPs at the lowest level.
Association with the performance of a discrete sys-
tem may define some MOPs. Others are continuous
performance variables such as sustained speed.
 3. Determine MOP value and hierarchy the goal is 32 knots. Pair-wise comparison is ac-
weighting factors. Use expert opinion and pair- complished for discrete values of speed at one knot
wise comparison to determine MOP value and the increments, and a value function is fit to these results
quantitative relationship between the OMOE and to calculate intermediate values. Again, the pair-
MOPs. wise comparison is structured to compare the rela-
Overall Mission
tive value of MOP options to achieve a partic ular
Effectiveness
OMOE MOE (Ordnance on target, etc.) in a specific sce-
SAG ARG Escort MCM Escort
nario.
SAG OMOE1 ARG OMOE2 MCM OMOE3
Value
1.000
Ordnance on Ordnance on # Ships Left # Landing Craft
Inland Targets Shore Targets w/Min Degradation Reached Beach
MOE1 MOE2 MOE3 MOE4
0.900
0.800
Mission Sustainability Mobility Vulnerability Susceptability
0.700
AAW-MOP1 Weapons-MOP9 Speed-MOP12 PSink/Hit - MOP15 IR Signiture - MOP17
ASUW-MOP2 Range-MOP10 Seakeeping-MOP13 Redundancy - MOP16 Acoustic Signature
0.600
ASW-MOP3 Duration-MOP11 Reliability - MOP14 - MOP18
C4I-MOP4 Hull RCS - MOP19 0.500
MCM-MOP5 Topside RCS - MOP20
NSFS-MOP6 0.400
SEW-MOP7
STK-MOP8 0.300
Figure 5. Notional Top Level OMOE Hierarchy 0.200
0.100
0.000
1.000 26 27 28 29 30 31 32
0.800 Speed (knots)
Value

32 cells 64 cells
Weapons Capacity
Figure 6. Discrete MOP Value Function
Figure 6 illustrates an example value index for
ship weapons capacity derived using pair-wise com-
parison. In this model weapons capacity is both a
discrete MOP, where it represents the performance
associated with this capacity, and a design parame-
ter. The metric for this MOP is the number of verti-
cal launch missile cells. The MOP threshold is 32
cells, and the MOP goal is 128 cells. Thresholds
represent the absolute minimum acceptable perform-
ance. Goals typically represent either a point of di-
minishing marginal value or a technology limitation.
The pair-wise comparison is structured to compare
the relative value of MOP options to achieve a par-
ticular MOE (Ordnance on shore target, etc.) in a
specific scenario.
Figure 7 illustrates an example value index for
ship sustained speed. Sustained speed is a continu-
ous MOP, not a design parameter. It is a function of
the ship design, primarily the hull form and installed
power. The threshold for this MOP is 26 knots, and
0.600
Figure 7. Continuous MOP Value Function
0.400
0.200
Once MOP value is determined for all MOPs,
0.000
pair-wise comparison is used to determine MOP and
128 cells MOE hierarchy weights. In this case the pair-wise
comparison is structured to compare the relative
value of achieving the goal in the first MOP or MOE
and only the threshold in the second, versus achie v-
ing only the threshold in the first MOP and the goal
in the second. This pair-wise comparison is accom-
plished at all levels of the hierarchy. A maximum ei-
genvalue approach is used to extract and quantify
average relative values and an inconsistency meas-
urement. MOP weights calculated for the Figure 5
notional hierarchy are shown in Figure 8. An
OMOE function, OMOE = g(MOP), is derived from
these weights and from the MOP value functions.
Ship Design Synthesis Model
The ship synthesis model used in this research is
based on a model originally developed by Reed
(1976). Reeds model was based on two earlier
codes, DD07 and CODESHIP (CNA, 1971).
Reeds model has been improved and updated at
MIT for over two decades by a long series of naval
officer students and faculty, and specifically for use
with a genetic algorithm (GA) by Shahak (1998). It
follows the basic process shown in Figure 9.

6
 MOP
MOP17-IR Signature

MOP18-Acoustic Signature

MOP16-Redundancy

MOP19-Hull RCS

MOP12-Sustained Speed

MOP11-Stores Duration

MOP8-STK

MOP20-Topside RCS

MOP15-CBR

MOP5-MCM

MOP13-Seakeeping

MOP10-Range

MOP7-SEW

MOP14-Reliability

MOP3-ASW

MOP9-Weapons

MOP4-C4I

MOP2-ASUW

MOP1-AAW

MOP6-NSFS

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

Weight
Figure 8. Measure of Performance (MOP) Weights

Input Design
Estimate Full
Load Weight
Calculate
Resistance
assess fitness and breed the next generation of ship
Start Principle
Parameters and Deckhouse
Volume
Characteristics
and Power variants.
No Balance requires that physical and functional
Feasible? Weight and Area and
constraints are satisfied. The ship must float. It
Tankage
Converge? Stability Volume must have adequate stability, volume, area, electric
Yes power, etc. It must provide required capabilities and
MOP's
satisfy minimum thresholds for performance. The
Risk Cost OMOE
ship synthesis model uses regression-based equations
Figure 9. Ship Synthesis Model Process for weight, volume, area and electric power. Resis-
tance is calculated using Gertler/Taylor Standard Se-
Most recently modules have been added to inter-
ries (Gertler, 1954). Cost is calculated using a modi-
face with a payload database, and calculate acquisi-
fied weight-based algorithm. LCC as defined for this
tion cost, seakeeping index and effectiveness. It has
analysis includes only follow-ship acquisition cost, life
also been updated to be more consistent with regres-
cycle fuel cost and life cycle manning cost.
sion-based equations for weight, area and electric
Seakeeping is assessed using the McCreight Index.
power as they have evolved in ASSET (DTMB,
1990). The current version is self-balancing, and
Cp Cx C L CBT CD10
written in FORTRAN 90. 0.61 0.82 80 2.9 11.1
In the GA application of this synthesis model, in- CRD Cmanning AAW ASUW ASW
put design parameters (genes) are specified in a ship 0.2 0.5 1 2 1
design matrix (chromosome). An example is shown C4I MCM NSFS SEW Weapons
in Figure 10. Design parameter descriptions are 1 4 1 1 1
Range Stores Shafts CPS ICR/GT
listed in Table 1. Specific payload systems with 3 2 2 1 1
weight, area and power requirements are associated
Figure 10. Design Parameter Chromosome
with each payload description. The ship is balanced
and resulting MOPs, OMOE, and Life Cycle Cost
(LCC) are calculated. The GA uses these results to

7
 Table 1. Design Parameter Descriptions
Design Parameter
1 - Prismatic Coefficient (Cp)
2 - Maximum Section Coeffi-
cient (Cx)
3 - Displacement to Length Ra-
tio (C)L)
4 - Beam to Draft Ratio (CBT)
5 - Length to Depth Ratio
(CD10)
6 - Raised Deck Ratio (CRD)
7 - Manning Factor (CManning)
8 - AAW Payload
9 - ASUW Payload
10 - ASW Payload
11 - C4I Payload
12 - MCM Payload
13 - NSFS Payload
14 - SEW Payload
15 - Weapons Capacity (VLS)
16 Range or fuel capacity
17 - Stores Duration
18 - Shafts
19 - CPS
20 - ICR or GT
Design Optimization
Ship design optimization is not a new concept,
but it poses difficult problems (Leopold, 1965,
Mandel, 1966 and Mandel, 1972). As discussed pre-
viously, ship design space is non-linear, very discon-
tinuous, and bounded by a variety of constraints and
thresholds. These attributes inhibit effective applic a-
Description tion of mature gradient-based optimization techniques
0.5-0.7; 20 increments
0.7-0.9; 20 increments
including Lagrange multipliers, steepest ascent meth-
ods, linear programming, non-linear programming and
60.0-90.0; 15increments dynamic programming.
In the simplified design problem presented in this
2.8-3.7; 9 increments
10.0-15.0; 10 increments
paper, each design variant requires 12.5 seconds on
a 200 MHz PC to balance and evaluate. An ex-
0.0-0.4; 4 increments haustive search would assess over ten trillion vari-
0.5-1.0; 5 increments ants requiring over 4 million years on this machine!
1 - Theater TBMD
Random search does not require a closed-form solu-
2 - Area TBMD
3 - Area Defense tion and has advantages of simplicity and insensitivity
4 - Limited Area Defense to discontinuities, but it still requires many iterations,
5 - Self Defense and is computationally impractical for a large design
1 - Long Range
2 - Medium Range
problem. Exponential random search improves the
3 - Short Range efficiency of random search, but also requires many
4 - Self Defense concept iterations. Genetic algorithms (GA) offer
1 - Area Domonance great promise to tackle this difficult problem.
2 - Adverse ASW Environ-
ment
Twenty-five years ago John Holland developed a
3 - Good ASW Environment genetic algorithm to abstract and explain the adaptive
4 - Torpedo Defense processes of natural systems for use in the design of
1 - Advanced artificial systems (Goldberg, 1989). Since that time,
2 - Current
1 - Limited Clearance
this algorithm has been applied with success to a
2 - Mine Recon wide range of problems. Genetic algorithms use
3 - Mine Avoidance models of natural selection, reproduction and muta-
4 - Limited Mine Advoidance
tion to improve a population of individuals or variants
1 - Advanced (VGAS,
NATACMS, ATWCS) based on the survival of the fittest, or in the case
2 - Full of Pareto Genetic Algorithms (PGAs), based on the
3 - Medium dominance and distribution of variants (Thomas,
4 - Minimum
1998). GAs are ideally suited to optimizing discon-
1 - Advanced
2 - Current tinuous and disjointed functions, and to optimization
1 - 128 cells where no closed-form function exists (or no mathe-
2 - 64 cells matical function at all, as with experimental data).
3 - 32 cells
The robustness of a particular GA depends on its ex-
1 - 10000 nm
2 - 7000 nm ploration and efficiency qualities. Exploration refers
3 - 5000 nm to its ability to master the design space and consis-
4 - 4000 nm tently identify the global optima. Efficiency refers to
1 - 60 days
2 - 45 days
the effort required to identify the global optima. Ro-
1 or 2 bustness implies an effective balance between these
0 (none) or 1 (full) qualities. Genetic algorithms are very robust relative
0 (ICR) or 1 (LM2500) to other methods.
Figure 11 illustrates the PGA process used in this
application. An initial population of 200 variants is
created by random selection of design variables
within the design space specified in Table 1. A
chromosome with 20 design parameters as in Figure
10 represents each variant. The ship variants de-
fined by these chromosomes are balanced, and
evaluated using the ship synthesis model resulting in
8
an assessment of feasibility and objective values
(OMOE and LCC) for each variant. Fitness indi-
cates a variants relative dominance in the population
based on OMOE and LCC. A Goldberg (1989)
ranking scheme is used. Variants are sorted into
layers of Pareto-dominance. Each layer contains all
variants that are dominant to subsequent layers.
Variants are sorted with the best variant in the high-
est layer getting a rank of one and the poorest in the
lowest layer a rank of 200 (population size). A crossover. A cut is made at a random location in the
geometrically decreasing probability of selection is
assigned to each variant based on its rank. Equiva-
lent variants (same dominance layer) are ordered
randomly within their layer, probabilities are aver-
aged for variants in the same layer, and the same
average value is ultimately assigned to each. Vari-
ants are penalized for infeasibility and for similarity to
other variants. This minimizes niches and duplicate
variants, and forces the selection to spread out over
the objective frontier.
Solution space
boundaries
Other input
No
Initialization
Method
par.in data file Yes
Random
initialization
structed with 200 segments representing each vari-
ant. The area of each segment is equal to the vari-
ant selection probability. Bakers (1987) selection
method is used. This method spins 200 (population
size) equally space markers once (vice spinning one
marker 200 times) to select 200 variants (some mul-
tiple times) for survival and reproduction.
Once a surviving population is selected, 25 per-
cent of these are chosen in pairs at random for
chromosomes of each pair. Design parameters be-
low the cut are swapped between the parents pro-
ducing new variants. A small percentage of individ-
ual design parameters (genes) in the selected vari-
ants are chosen randomly to mutate. In mutation, the
value of a single design parameter is replaced with a
new value chosen at random. After these operations
are completed, new variants in the new population
are evaluated and the process cycles until conver-
gence. Each cycle defines a new generation.
Compared to other optimization techniques, ge-
netic algorithms are particularly well suited for gen-
erating a Pareto-optimal frontier because they im-
prove the fitness of a population of concepts simulta-
Forced neously. By penalizing for niching, this population
spreads out over non-dominated values of the obje c-
DDG_forced
data file
tive attributes, and ultimately defines the Pareto-
optimal frontier.

gen=1,gen_max
Surface Combatant Design
Ship Synthesis
A simple surface-combatant concept design is
Self-balanced Model used to evaluate and demonstrate this design meth-
Yes No odology. A Mission Need Statement, Required Op-
Penalty? Feasible?
erational Capabilities (ROCs) and mission scenarios
No Yes
Feasibility Penalty were developed for a notional Land Attack Surface
Combatant. A group of warfare experts consisting
Sort into Dominance Layers / Niche
Penalty
of MIT naval officer students and faculty were
briefed on this mission need, and an effectiveness hi-
Selection
erarchy similar to Figure 5 was developed. Meas-
Crossover ures of Effectiveness were identified for each of
three critical mission scenarios: 1) Surface Action
Mutation Group (SAG); 2) Amphibious Readiness Group
(ARG) escort; and 3) Mine Counter Measures
No
Convergence? (MCM) Group escort. Ship MOPs were identified
Yes
for each MOE. Goals and thresholds were set for all
Report
and end MOEs and MOPs. The AHP was applied using
Figure 11. Pareto Genetic Algorithm (PGA) Process questionnaires and group discussion. AHP results
were used to build an OMOE function. Typical
Once selection probabilities have been assigned, MOP value functions and MOP weights are shown
selection can proceed. A roulette wheel is con- in Figures 6 through 8.

9
 Once the OMOE function was determined, criti- achieve various levels of warfare area performance,
cal design parameters to be varied were selected, ship endurance, number of shafts, use of a Colle ctive
and structured as a GA chromosome. These DPs Protection System (CPS), propulsion gas turbine se-
included hull form parameters, payload packages to lection, and a manpower reduction factor.
Table 2. Select Non-dominated Feasible Variants
Ship A Ship B Ship C Ship D
LBP (ft) 422.8 440.4 520.9 588.5
Beam (ft) 52.5 59.9 55.5 75.0
D10 (ft) 36.8 40.0 35.9 52.6
Draft (ft) 17.5 21.2 19.8 24.2
Displacement (lton) 5524.2 6713.8 8899.2 16612
Shafts 1 1 2 2
Range (nm) 4000 5000 7000 10000
Sustained Speed 27.1 27.0 31.0 29.2
(knt)
GM/B .108 .129 .101 .111
Generators/kW 3/3000 3/3000 3/3000 7/3000
CPS no yes full full
AAW RAM,SPS- ESSM,SM,SPS-49,X-Band ra- ESSM,SM,SPY-1D, X- ESSM,SM,SPY-1D, X and S Band
49,CIWS,SDS dar, Mk92 MFCS Band radar,Mk99 GMFCS radar, GMFCS
ASUW 5/54 w/ERGM,GFCS Harpoon, 5/54 Harpoon,AN/SWG- TASM/TMMM,
w/ERGM,GFCS 1,VGAS,GFCS ATWCS,VGAS,GFCS
ASW 1.5m sonar, SSTD, 5m sonar(passive), SSTD, 5m sonar, SSTD, LAMPS 5m sonar, SSTD, LAMPS MK3,
helo haven, NIXIE, SVTT,helo ha- MKIII,NIXIE,SVTT,VL NIXIE,SVTT,VLA,SQQ-89,
SVTT,NIXIE ven,VLA A LBVDS
C4I Baseline Baseline CEC,JTIDS, digital CEC, JTIDS, digital comm, TA-
comm, T A- DIX/TACINTEL
DIX/TACINTEL
MCM Degaussing Degaussing Mine avoidance sonar, Mine avoidance sonar, Remote
degaussing Minehunting System, degaussing
NSFS N-ATACMS, 5/54 N-ATACMS, 5/54 VGAS, N-ATACMS, VGAS, N-ATACMS, ATWCS
w/ERGM,GFCS w/ERGM,GFCS ATWCS
STK TWCS,TLAMs,UAVs TWCS,TLAMs,UAVs ATWCS,TLAMs,UAVs ATWCS,TLAMs,UAVs
SEW SLQ-32V2, DLS SLQ32V2,DLS AIEWS,DLS AIEWS,DLS
VLS cells 32 64 64 128
Hello hangar / helos 0 0 Yes/2 Yes/2
Crew 108 120 184 266
Follow ship Acquisi- 547.6 656.6 888.6 1242.3
tion Cost ($M)
LCC ($M) 628.3 943.5 1319.7 1867.5
OMOE 0.2296 .4627 .7523 .9562

OMOE
Ship manning was calculated based on historical
1 data and reduced by a manpower reduction factor.
Ship C Ship D
0.9 Weight and cost for automation was increased con-
0.8
Ship B
sistent with the manpower reduction. Weight, area
0.7
0.6
and power were specified for each payload package.
0.5
Ship A Chromosome DPs are described in Table 1. Other
0.4 DPs were held constant for all designs.
0.3 Gen 50
A PGA search was completed for the design
0.2 Gen 40 space specified in Table 1 using OMOE and LCC as
0.1 Gen 1 objective attributes. The optimization was run for 50
0 generations, taking 26 hours on a 200 MHz PC.
0.5 1 1.5 2 2.5 3
LCC was defined to include acquisition cost, dis-
Life Cycle Cost ($B)
counted life cycle fuel cost and discounted life cycle
Figure 12. OMOE/LCC Frontier manpower cost. Results of this search are pre-

10
sented in Figure 12. The data points represent LCC
and OMOE values for a sample of feasible variants
in generations 1, 40 and 50. Generation 1 is a ran-
dom selection of design parameters. Convergence to
a non-dominated frontier can be seen in the evolution
from Generation 1 to Generation 40 and finally to
Generation 50. Generation 50 results approximate
the design frontier. Many of these points are non-
dominated.
Ships A, B, C and D are non-dominated and rep-
resent knees in the curve. Characteristics for
these ships are listed in Table 2. Variants between
Ships C and D on the frontier have 2 shafts. Vari-
ants below Ship B on the frontier have one shaft.
Between Ships B and C, there is a mix of one and
two shaft ships. Ship D is the feasible ship with the
highest OMOE.
Although non-dominated, none of these ships can
be identified as the best. Selection of the pre-
ferred design is up to the customer, but Figure 12
provides the customer with important information to
make this selection: 1) the engineer can assure the
customer with confidence that non-dominated vari-
ants have been identified; 2) the non-dominated
frontier provides a perspective on the design space;
and 3) some variants stand out as providing good
value given a range of acceptable cost. In this ex-
ample, Ships A, B, C and D are noteworthy.
A discussion with the customer might consider
the following:
Ship A represents a low-end alternative. It
has good performance in Naval Surface Fire Support
(NSFS) with other MOPs at threshold values. It
represents the best alternative if acquisition cost is
limited to $500-600M. Most war fighters would not
be impressed, but it is the best for the money.
Ship D is the high-end variant. It achieves
goal performance in most MOPs. The big cost
driver for ships to the right of Ship C is the addition
of an S-band radar. This radar has a large ship im-
pact due to its power (seven 3000 kW generators)
and cooling requirements. Ship D would be an ex-
cellent candidate for an Integrated Power System
(IPS), an option which is not included in the present
model. As a mechanical drive ship, it is only worth
the money if full TBMD capability is essential.
Ship B is the most effective of the single shaft
ships at a reasonable price. Its AAW and ASW sys-
tems are much more capable than Ship A and the
11
acquisition cost is still low. It is an excellent choice
for a reasonable low-end capability.
Ship C stands out as a particularly sharp knee
in the curve. It has excellent capability in all areas
and the price is comparable to a number of variants
with much less effectiveness. It is a Best Buy,
and if the acquisition cost is manageable, this would
be an excellent choice.
Conclusions
It is estimated that more than 80 percent of a
naval ships ultimate acquisition cost is locked in dur-
ing concept design. For a class of ships, this means
tens of billions of dollars. An ad hoc process for
making these critical design decisions is not ade-
quate. Figure 11 appears to be a simple and some-
what intuitive result, but it is not, and its implic ations
are revolutionary. Without this kind of information,
we cannot make responsible decisions.
Key elements addressed by this methodology
are:
It provides a practical method for the ship
designer to calculate an Overall Measure
of Effectiveness (OMOE) which repre-
sents customer requirements and relates
ship measures of performance (MOPs) to
mission effectiveness. This is an essential
prerequisite to a disciplined search of de-
sign parameters.
It includes an efficient method to search
design space for non-dominated concepts.
It provides a consistent format for present-
ing and trading off a manageable set of
dissimilar objective attributes (effective-
ness, cost, and risk).
Essential future work includes:
Integrate with a design database and prod-
uct model.
Extend to later stages of design using a
dynamic landscape PGA.
Validate the AHP/MAVT approach to de-
fine and calculate an OMOE.
Integrate with a process/shipbuilding
model.
Extend to include risk as a third objective
attribute.
 Evaluate existing ship designs using this
methodology.
Compare the GA performance to other op-
timization techniques.
The methodology described in this paper does
not replace imagination and experience. It provides
a practical tool to manage a complex total-system
problem that cannot be managed by experience and
intuition alone. It represents essential change in how
we do naval ship concept design.
References
Baker, J.E. (1987), Reducing Bias and Inefficiency
in the Selection Algorithm, pp. 14-21, Proceed-
ings of the Second International Conference on
Genetic Algorithms, Hillsdale, NJ.
Belton, V. (1986), A comparison of the analytic hi-
erarchy process and a simple multi-attribute value
function, European Journal of Operational Re-
search.
DTMB (1990), ASSET/MONOSC User Manual,
Version 3.3, DTMB, Carderock Division, Naval
Surface Warfare Center.
Gertler, Morton (1954), A Reanalysis of the Original
Test Data for the Taylor Series, Report 806,
Navy Department, DTMB.
Goldberg, D.E. (1989), Genetic Algorithms in
Search, Optimization and Machine Learning,
Reading: Addison-Wesley Publishing Company,
Inc..
Keeney, R.L. and Raiffa, H. (1976), Decisions with
Multiple Objectives: Preferences and Value
Tradeoffs, John Wiley and Sons, New York.
Kramer, R. et. al. (1996), "Virtual Notational Ship
Simulation Supporting SC-21 Design", Modeling,
Simulation and Virtual Prototyping Conference,
ASNE.
Hockberger, W.A. (1996), "Total System Ship De-
sign in a Supersystem Framework", Naval Engi-
neers Journal.
Leopold, R. (1965), Mathematical Optimization
Methods Applied to Ship Design, MIT Depart-
ment of Naval Architecture and Marine Engineer-
ing, Report 65-8.
Mandel, P., and Leopold, R. (1966), "Optimization
Methods Applied to Ship Design", SNAME Trans-
actions, Vol. 74.
12
Mandel, P., and Chryssostomidis, C. (1972), "A De-
sign Methodology for Ships and Other Complex
Systems", Philosphical Transactions, Royal Society
of London, A273.
Saaty, T.L. (1996), The Analytic Hierarchy Proc-
ess, RWS Publications, Pittsburgh.
Shahak, S. (1998), Naval Ship Concept Design: an
Evolutionary Approach, Master of Science The-
sis, MIT Department of Ocean Engineering.
Suh, N.P. (1990), The Principles of Design, Oxford
University Press, New York.
Thomas, Mark (1998), A Pareto Frontier for Full
Stern Submarines via Genetic Algorithm PhD
Thesis, MIT Department of Ocean Engineering.
Tibbitts, B. and Keane, R.G. (1995), "Making Design
Everybody's Job", Naval Engineers Journal.
Dr. Alan Brown, CAPT, USN (ret) is currently
Professor, Department of Aeronautics and Ocean
Engineering, Virginia Tech. He was Professor of
Naval Architecture, and directed the Naval Con-
struction and Engineering Program at MIT from
1993 to 1997. As an Engineering Duty Officer he
served in ships, fleet staffs, shipyards, NAVSEA
and OPNAV. While at MIT and since retirement
he has served as technical advisor to US delega-
tions at the International Maritime Organization
(IMO) in tanker design, oil outflow, intact stability,
damaged stability and tanker risk. He is chairman
of the SNAME Ad Hoc Panel on Structural De-
sign and Response in Collision and Grounding, a
member of the SNAME Ship Design Committee
and SNAME Panel O-44, Marine Safety and Pol-
lution Prevention. He is Northeast Regional Vice
President of SNAME, a member of the ASNE
Council and Past Chairman of the New England
Section of SNAME. He received a PhD in Ma-
rine Engineering from MIT in 1986.
LCDR Mark Thomas, USN graduated from Okla-
homa State University with a bachelor's degree in
electrical engineering. His first Navy assignment
was aboard USS David R. Ray (DD-971) where
he served as First Lieutenant, Gunnery Officer,
and Auxiliaries Officer. After four years, he re-
turned to Surface Warfare Officers School
(SWOS) Coronado as an instructor and course di-
rector for the DD-963 Engineering Officer of the
Watch (EOOW) School. He transferred to the ED
community in 1991, and served two years as a
main propulsion inspector at the Pacific Board of
Inspection and Survey. He reported to MIT in
May 1993, and received the degrees of Naval En-
gineer and Master of Science in Electrical Engi-
neering and Computer Science in June 1996. He
then began a two-year period of doctoral research
and received the degree of Doctor of Philosophy in
hydrodynamics in June 1998. LCDR Thomas' next
permanent duty station will be at the Supervisor of
Shipbuilding, Pascagoula MS.

13